LTX-2.5 in ComfyUI on a 20 GB GPU: Draft → Learned x2 Upscale → Refine (What Actually Matters)

Short answer: on a 20 GB card, run LTX-2.5 as three stages, not one: Draft at half the final resolution, 8 steps, distilled model at cfg 1.0. Upscale the latent x2 with LTX's learned spatial upscaler